Chicken Nuggets

Inspiration

We selected the challenge that was provided by Cyclica since it looks unique among others as all of us were intrigued by testing protein with our own machine-learning model. It was a novel field that we have never experienced before, which inspires us to run into this challenge.

What it does

It is a classification model to predict 'drug binding' or 'non-drug binding' for any query residue.

How we built it

To brainstorm on how we proceed with our projects, first, we used pandas to read data stored in CSV files and went through the preprocessing by dropping unnecessary columns and replacing zero values with a median. Furthermore, the non-integer data was converted to an integer value by SKlearn. After walking through the data, we decided to utilize logistic regression to train the model as we were working on a binary outcome variable and one or more independent variables.

Challenges we ran into

At first, we were thrilled since it only took a few hours to achieve 95% accuracy. However, we soon realized that this was due to the model producing only false results and there was an enormous bias in our training sets. To fix this, we went through a hard time. We suffered to guess what features are necessary for the model.

Accomplishments that we're proud of

Overall, our accuracy was 69%, and the F1 score was 72%. Our precision score was 67%, our recall score was 77%, and our ROC AUC score was 69%. While some people might find this disappointing, we are incredibly proud of what we accomplished.

What we learned

This was our first-ever machine-learning project, and we learned a lot along the way. We faced several challenges, but we persevered and came out on top. We learned that accuracy isn't everything and that there are several other metrics to consider when evaluating a model's performance.

What's next for Chicken Nuggets

For now, we are extremely satisfied with what we have achieved in this short period of time. On the other hand, we are planning to participate in other Datathons since it was a memorable experience with the CxC hackathon.